diff options
author | Roman Lebedev <lebedev.ri@gmail.com> | 2018-09-20 07:54:49 +0000 |
---|---|---|
committer | Roman Lebedev <lebedev.ri@gmail.com> | 2018-09-20 07:54:49 +0000 |
commit | 38c25ace538adc572e8d26a90a6f149419ff41eb (patch) | |
tree | 12db593461dc61f2b728f95f142e357086b413ac /llvm/test/CodeGen/X86/extract-lowbits.ll | |
parent | 439c29f5eada56ac41de5059198265aa15f53e7a (diff) | |
download | bcm5719-llvm-38c25ace538adc572e8d26a90a6f149419ff41eb.tar.gz bcm5719-llvm-38c25ace538adc572e8d26a90a6f149419ff41eb.zip |
[NFC][x86][AArch64] Add BEXTR-like test patterns.
Summary: Also, adjust the check prefixes so that we actually get to check the BMI1-only-case.
Reviewers: craig.topper, RKSimon, spatel, javed.absar
Reviewed By: RKSimon
Subscribers: kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D48490
llvm-svn: 342623
Diffstat (limited to 'llvm/test/CodeGen/X86/extract-lowbits.ll')
-rw-r--r-- | llvm/test/CodeGen/X86/extract-lowbits.ll | 1021 |
1 files changed, 1019 insertions, 2 deletions
diff --git a/llvm/test/CodeGen/X86/extract-lowbits.ll b/llvm/test/CodeGen/X86/extract-lowbits.ll index ee715430530..4af130cd825 100644 --- a/llvm/test/CodeGen/X86/extract-lowbits.ll +++ b/llvm/test/CodeGen/X86/extract-lowbits.ll @@ -1,11 +1,11 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI,X86-NOBMI -; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1 +; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1NOTBM,X86-BMI1NOTBM ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1TBM,X86-BMI1TBM ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1TBM,X86-BMI1TBM,BMI1TBMBMI2,X86-BMI1TBMBMI2 ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1NOTBMBMI2,X86-BMI1NOTBMBMI2 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI,X64-NOBMI -; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1 +; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1NOTBM,X64-BMI1NOTBM ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1TBM,X64-BMI1TBM ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1TBM,X64-BMI1TBM,BMI1TBMBMI2,X64-BMI1TBMBMI2 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1NOTBMBMI2,X64-BMI1NOTBMBMI2 @@ -37,6 +37,15 @@ define i32 @bzhi32_a0(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_a0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: decl %eax +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_a0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -53,6 +62,16 @@ define i32 @bzhi32_a0(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl %edi, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_a0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: decl %eax +; X64-BMI1NOTBM-NEXT: andl %edi, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_a0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -73,6 +92,15 @@ define i32 @bzhi32_a1_indexzext(i32 %val, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_a1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: decl %eax +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_a1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -89,6 +117,16 @@ define i32 @bzhi32_a1_indexzext(i32 %val, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl %edi, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_a1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: decl %eax +; X64-BMI1NOTBM-NEXT: andl %edi, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_a1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -111,6 +149,16 @@ define i32 @bzhi32_a2_load(i32* %w, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl (%edx), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_a2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: decl %eax +; X86-BMI1NOTBM-NEXT: andl (%edx), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_a2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -128,6 +176,16 @@ define i32 @bzhi32_a2_load(i32* %w, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl (%rdi), %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_a2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: decl %eax +; X64-BMI1NOTBM-NEXT: andl (%rdi), %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_a2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -150,6 +208,16 @@ define i32 @bzhi32_a3_load_indexzext(i32* %w, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl (%edx), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_a3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: decl %eax +; X86-BMI1NOTBM-NEXT: andl (%edx), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_a3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -167,6 +235,16 @@ define i32 @bzhi32_a3_load_indexzext(i32* %w, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl (%rdi), %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_a3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: decl %eax +; X64-BMI1NOTBM-NEXT: andl (%rdi), %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_a3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -189,6 +267,15 @@ define i32 @bzhi32_a4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_a4_commutative: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: decl %eax +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_a4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -205,6 +292,16 @@ define i32 @bzhi32_a4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl %edi, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_a4_commutative: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: decl %eax +; X64-BMI1NOTBM-NEXT: andl %edi, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_a4_commutative: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -237,6 +334,25 @@ define i64 @bzhi64_a0(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_a0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB5_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: xorl %eax, %eax +; X86-BMI1NOTBM-NEXT: .LBB5_2: +; X86-BMI1NOTBM-NEXT: addl $-1, %eax +; X86-BMI1NOTBM-NEXT: adcl $-1, %edx +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_a0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl @@ -266,6 +382,16 @@ define i64 @bzhi64_a0(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_a0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: decq %rax +; X64-BMI1NOTBM-NEXT: andq %rdi, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_a0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -296,6 +422,25 @@ define i64 @bzhi64_a1_indexzext(i64 %val, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_a1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB6_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: xorl %eax, %eax +; X86-BMI1NOTBM-NEXT: .LBB6_2: +; X86-BMI1NOTBM-NEXT: addl $-1, %eax +; X86-BMI1NOTBM-NEXT: adcl $-1, %edx +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_a1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl @@ -325,6 +470,16 @@ define i64 @bzhi64_a1_indexzext(i64 %val, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_a1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: decq %rax +; X64-BMI1NOTBM-NEXT: andq %rdi, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_a1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -360,6 +515,28 @@ define i64 @bzhi64_a2_load(i64* %w, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %esi ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_a2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB7_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: xorl %eax, %eax +; X86-BMI1NOTBM-NEXT: .LBB7_2: +; X86-BMI1NOTBM-NEXT: addl $-1, %eax +; X86-BMI1NOTBM-NEXT: adcl $-1, %edx +; X86-BMI1NOTBM-NEXT: andl 4(%esi), %edx +; X86-BMI1NOTBM-NEXT: andl (%esi), %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_a2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -392,6 +569,16 @@ define i64 @bzhi64_a2_load(i64* %w, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq (%rdi), %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_a2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: decq %rax +; X64-BMI1NOTBM-NEXT: andq (%rdi), %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_a2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax @@ -426,6 +613,28 @@ define i64 @bzhi64_a3_load_indexzext(i64* %w, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %esi ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_a3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB8_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: xorl %eax, %eax +; X86-BMI1NOTBM-NEXT: .LBB8_2: +; X86-BMI1NOTBM-NEXT: addl $-1, %eax +; X86-BMI1NOTBM-NEXT: adcl $-1, %edx +; X86-BMI1NOTBM-NEXT: andl 4(%esi), %edx +; X86-BMI1NOTBM-NEXT: andl (%esi), %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_a3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -458,6 +667,16 @@ define i64 @bzhi64_a3_load_indexzext(i64* %w, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq (%rdi), %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_a3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: decq %rax +; X64-BMI1NOTBM-NEXT: andq (%rdi), %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_a3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -491,6 +710,25 @@ define i64 @bzhi64_a4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_a4_commutative: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $1, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB9_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: xorl %eax, %eax +; X86-BMI1NOTBM-NEXT: .LBB9_2: +; X86-BMI1NOTBM-NEXT: addl $-1, %eax +; X86-BMI1NOTBM-NEXT: adcl $-1, %edx +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_a4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl @@ -520,6 +758,16 @@ define i64 @bzhi64_a4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_a4_commutative: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movl $1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: decq %rax +; X64-BMI1NOTBM-NEXT: andq %rdi, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_a4_commutative: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -544,6 +792,14 @@ define i32 @bzhi32_b0(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_b0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %eax, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_b0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -560,6 +816,15 @@ define i32 @bzhi32_b0(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl %edi, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_b0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $-1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: andnl %edi, %eax, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_b0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -580,6 +845,14 @@ define i32 @bzhi32_b1_indexzext(i32 %val, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_b1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %eax, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_b1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -596,6 +869,15 @@ define i32 @bzhi32_b1_indexzext(i32 %val, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl %edi, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_b1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $-1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: andnl %edi, %eax, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_b1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -618,6 +900,15 @@ define i32 @bzhi32_b2_load(i32* %w, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl (%edx), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_b2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %edx +; X86-BMI1NOTBM-NEXT: andnl (%eax), %edx, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_b2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -635,6 +926,15 @@ define i32 @bzhi32_b2_load(i32* %w, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl (%rdi), %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_b2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $-1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: andnl (%rdi), %eax, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_b2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -657,6 +957,15 @@ define i32 @bzhi32_b3_load_indexzext(i32* %w, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl (%edx), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_b3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shll %cl, %edx +; X86-BMI1NOTBM-NEXT: andnl (%eax), %edx, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -674,6 +983,15 @@ define i32 @bzhi32_b3_load_indexzext(i32* %w, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl (%rdi), %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_b3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $-1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: andnl (%rdi), %eax, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -696,6 +1014,14 @@ define i32 @bzhi32_b4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_b4_commutative: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %eax, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_b4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -712,6 +1038,15 @@ define i32 @bzhi32_b4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andl %edi, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_b4_commutative: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl $-1, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: andnl %edi, %eax, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_b4_commutative: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -744,6 +1079,25 @@ define i64 @bzhi64_b0(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_b0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB15_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %esi, %eax +; X86-BMI1NOTBM-NEXT: xorl %esi, %esi +; X86-BMI1NOTBM-NEXT: .LBB15_2: +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %eax, %edx +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %esi, %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_b0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -772,6 +1126,15 @@ define i64 @bzhi64_b0(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_b0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq $-1, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: andnq %rdi, %rax, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_b0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -802,6 +1165,25 @@ define i64 @bzhi64_b1_indexzext(i64 %val, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_b1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB16_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %esi, %eax +; X86-BMI1NOTBM-NEXT: xorl %esi, %esi +; X86-BMI1NOTBM-NEXT: .LBB16_2: +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %eax, %edx +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %esi, %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_b1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -830,6 +1212,15 @@ define i64 @bzhi64_b1_indexzext(i64 %val, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_b1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movq $-1, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: andnq %rdi, %rax, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_b1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -865,6 +1256,26 @@ define i64 @bzhi64_b2_load(i64* %w, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %esi ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_b2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: movl $-1, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %edx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB17_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %esi, %edx +; X86-BMI1NOTBM-NEXT: xorl %esi, %esi +; X86-BMI1NOTBM-NEXT: .LBB17_2: +; X86-BMI1NOTBM-NEXT: andnl 4(%eax), %edx, %edx +; X86-BMI1NOTBM-NEXT: andnl (%eax), %esi, %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_b2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -894,6 +1305,15 @@ define i64 @bzhi64_b2_load(i64* %w, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq (%rdi), %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_b2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq $-1, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: andnq (%rdi), %rax, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_b2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax @@ -928,6 +1348,26 @@ define i64 @bzhi64_b3_load_indexzext(i64* %w, i8 zeroext %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %esi ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_b3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: movl $-1, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %edx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB18_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %esi, %edx +; X86-BMI1NOTBM-NEXT: xorl %esi, %esi +; X86-BMI1NOTBM-NEXT: .LBB18_2: +; X86-BMI1NOTBM-NEXT: andnl 4(%eax), %edx, %edx +; X86-BMI1NOTBM-NEXT: andnl (%eax), %esi, %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_b3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -957,6 +1397,15 @@ define i64 @bzhi64_b3_load_indexzext(i64* %w, i8 zeroext %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq (%rdi), %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_b3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movq $-1, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: andnq (%rdi), %rax, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_b3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -990,6 +1439,25 @@ define i64 @bzhi64_b4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_b4_commutative: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB19_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %esi, %eax +; X86-BMI1NOTBM-NEXT: xorl %esi, %esi +; X86-BMI1NOTBM-NEXT: .LBB19_2: +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %eax, %edx +; X86-BMI1NOTBM-NEXT: andnl {{[0-9]+}}(%esp), %esi, %eax +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_b4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -1018,6 +1486,15 @@ define i64 @bzhi64_b4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_b4_commutative: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq $-1, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: andnq %rdi, %rax, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_b4_commutative: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -1043,6 +1520,16 @@ define i32 @bzhi32_c0(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_c0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_c0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1059,6 +1546,16 @@ define i32 @bzhi32_c0(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_c0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl %edi, %eax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_c0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -1080,6 +1577,16 @@ define i32 @bzhi32_c1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_c1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_c1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -1096,6 +1603,16 @@ define i32 @bzhi32_c1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_c1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl %edi, %eax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_c1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -1119,6 +1636,17 @@ define i32 @bzhi32_c2_load(i32* %w, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_c2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl (%eax), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_c2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1136,6 +1664,16 @@ define i32 @bzhi32_c2_load(i32* %w, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_c2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl (%rdi), %eax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_c2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -1159,6 +1697,17 @@ define i32 @bzhi32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_c3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl (%eax), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1176,6 +1725,16 @@ define i32 @bzhi32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_c3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl (%rdi), %eax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -1199,6 +1758,16 @@ define i32 @bzhi32_c4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_c4_commutative: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_c4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1215,6 +1784,16 @@ define i32 @bzhi32_c4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_c4_commutative: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl %edi, %eax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_c4_commutative: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -1246,6 +1825,24 @@ define i64 @bzhi64_c0(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %edx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_c0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl $64, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shrl %cl, %edx +; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB25_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %edx, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: .LBB25_2: +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_c0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl $64, %ecx @@ -1273,6 +1870,16 @@ define i64 @bzhi64_c0(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_c0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq %rdi, %rax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_c0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -1302,6 +1909,24 @@ define i64 @bzhi64_c1_indexzext(i64 %val, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %edx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_c1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shrl %cl, %edx +; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB26_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %edx, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: .LBB26_2: +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_c1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb $64, %cl @@ -1329,6 +1954,16 @@ define i64 @bzhi64_c1_indexzext(i64 %val, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_c1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movq %rdi, %rax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_c1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -1363,6 +1998,27 @@ define i64 @bzhi64_c2_load(i64* %w, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %esi ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_c2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1NOTBM-NEXT: movl $64, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shrl %cl, %edx +; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB27_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %edx, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: .LBB27_2: +; X86-BMI1NOTBM-NEXT: andl (%esi), %eax +; X86-BMI1NOTBM-NEXT: andl 4(%esi), %edx +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_c2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -1393,6 +2049,16 @@ define i64 @bzhi64_c2_load(i64* %w, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_c2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq (%rdi), %rax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_c2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax @@ -1426,6 +2092,27 @@ define i64 @bzhi64_c3_load_indexzext(i64* %w, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %esi ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_c3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shrl %cl, %edx +; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB28_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %edx, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: .LBB28_2: +; X86-BMI1NOTBM-NEXT: andl (%esi), %eax +; X86-BMI1NOTBM-NEXT: andl 4(%esi), %edx +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi @@ -1456,6 +2143,16 @@ define i64 @bzhi64_c3_load_indexzext(i64* %w, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_c3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movq (%rdi), %rax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -1488,6 +2185,24 @@ define i64 @bzhi64_c4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %edx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_c4_commutative: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl $64, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movl $-1, %eax +; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: shrl %cl, %edx +; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: je .LBB29_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %edx, %eax +; X86-BMI1NOTBM-NEXT: xorl %edx, %edx +; X86-BMI1NOTBM-NEXT: .LBB29_2: +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: andl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_c4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl $64, %ecx @@ -1515,6 +2230,16 @@ define i64 @bzhi64_c4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_c4_commutative: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq %rdi, %rax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_c4_commutative: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -1540,6 +2265,16 @@ define i32 @bzhi32_d0(i32 %val, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_d0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_d0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1556,6 +2291,16 @@ define i32 @bzhi32_d0(i32 %val, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_d0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl %edi, %eax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_d0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -1577,6 +2322,16 @@ define i32 @bzhi32_d1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_d1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_d1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al @@ -1593,6 +2348,16 @@ define i32 @bzhi32_d1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_d1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl %edi, %eax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_d1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax @@ -1616,6 +2381,17 @@ define i32 @bzhi32_d2_load(i32* %w, i32 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_d2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl (%eax), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_d2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1633,6 +2409,16 @@ define i32 @bzhi32_d2_load(i32* %w, i32 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_d2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl (%rdi), %eax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_d2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -1656,6 +2442,17 @@ define i32 @bzhi32_d3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi32_d3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl (%eax), %eax +; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: shll %cl, %eax +; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax @@ -1673,6 +2470,16 @@ define i32 @bzhi32_d3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrl %cl, %eax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi32_d3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movl (%rdi), %eax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shll %cl, %eax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrl %cl, %eax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax @@ -1727,6 +2534,45 @@ define i64 @bzhi64_d0(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %ebx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_d0: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %ebx +; X86-BMI1NOTBM-NEXT: pushl %edi +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl $64, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movl %edx, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl %esi, %edi +; X86-BMI1NOTBM-NEXT: jne .LBB34_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edi +; X86-BMI1NOTBM-NEXT: .LBB34_2: +; X86-BMI1NOTBM-NEXT: movl %edi, %eax +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl $0, %edx +; X86-BMI1NOTBM-NEXT: jne .LBB34_4 +; X86-BMI1NOTBM-NEXT: # %bb.3: +; X86-BMI1NOTBM-NEXT: movl %esi, %ebx +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: .LBB34_4: +; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: jne .LBB34_6 +; X86-BMI1NOTBM-NEXT: # %bb.5: +; X86-BMI1NOTBM-NEXT: movl %ebx, %eax +; X86-BMI1NOTBM-NEXT: .LBB34_6: +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: popl %edi +; X86-BMI1NOTBM-NEXT: popl %ebx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_d0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %edi @@ -1769,6 +2615,16 @@ define i64 @bzhi64_d0(i64 %val, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_d0: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq %rdi, %rax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_d0: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax @@ -1819,6 +2675,45 @@ define i64 @bzhi64_d1_indexzext(i64 %val, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %ebx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_d1_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %ebx +; X86-BMI1NOTBM-NEXT: pushl %edi +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl %edx, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl %esi, %edi +; X86-BMI1NOTBM-NEXT: jne .LBB35_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edi +; X86-BMI1NOTBM-NEXT: .LBB35_2: +; X86-BMI1NOTBM-NEXT: movl %edi, %eax +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl $0, %edx +; X86-BMI1NOTBM-NEXT: jne .LBB35_4 +; X86-BMI1NOTBM-NEXT: # %bb.3: +; X86-BMI1NOTBM-NEXT: movl %esi, %ebx +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: .LBB35_4: +; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: jne .LBB35_6 +; X86-BMI1NOTBM-NEXT: # %bb.5: +; X86-BMI1NOTBM-NEXT: movl %ebx, %eax +; X86-BMI1NOTBM-NEXT: .LBB35_6: +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: popl %edi +; X86-BMI1NOTBM-NEXT: popl %ebx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_d1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %edi @@ -1861,6 +2756,16 @@ define i64 @bzhi64_d1_indexzext(i64 %val, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_d1_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movq %rdi, %rax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_d1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -1914,6 +2819,46 @@ define i64 @bzhi64_d2_load(i64* %w, i64 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %ebx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_d2_load: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %ebx +; X86-BMI1NOTBM-NEXT: pushl %edi +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl (%eax), %edx +; X86-BMI1NOTBM-NEXT: movl 4(%eax), %eax +; X86-BMI1NOTBM-NEXT: movl $64, %ecx +; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movl %edx, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl %esi, %edi +; X86-BMI1NOTBM-NEXT: jne .LBB36_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edi +; X86-BMI1NOTBM-NEXT: .LBB36_2: +; X86-BMI1NOTBM-NEXT: movl %edi, %eax +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl $0, %edx +; X86-BMI1NOTBM-NEXT: jne .LBB36_4 +; X86-BMI1NOTBM-NEXT: # %bb.3: +; X86-BMI1NOTBM-NEXT: movl %esi, %ebx +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: .LBB36_4: +; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: jne .LBB36_6 +; X86-BMI1NOTBM-NEXT: # %bb.5: +; X86-BMI1NOTBM-NEXT: movl %ebx, %eax +; X86-BMI1NOTBM-NEXT: .LBB36_6: +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: popl %edi +; X86-BMI1NOTBM-NEXT: popl %ebx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_d2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %edi @@ -1957,6 +2902,16 @@ define i64 @bzhi64_d2_load(i64* %w, i64 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_d2_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx +; X64-BMI1NOTBM-NEXT: movq (%rdi), %rax +; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_d2_load: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax @@ -2009,6 +2964,46 @@ define i64 @bzhi64_d3_load_indexzext(i64* %w, i8 %numlowbits) nounwind { ; X86-NOBMI-NEXT: popl %ebx ; X86-NOBMI-NEXT: retl ; +; X86-BMI1NOTBM-LABEL: bzhi64_d3_load_indexzext: +; X86-BMI1NOTBM: # %bb.0: +; X86-BMI1NOTBM-NEXT: pushl %ebx +; X86-BMI1NOTBM-NEXT: pushl %edi +; X86-BMI1NOTBM-NEXT: pushl %esi +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movl (%eax), %edx +; X86-BMI1NOTBM-NEXT: movl 4(%eax), %eax +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl %edx, %esi +; X86-BMI1NOTBM-NEXT: shll %cl, %esi +; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %eax +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl %esi, %edi +; X86-BMI1NOTBM-NEXT: jne .LBB37_2 +; X86-BMI1NOTBM-NEXT: # %bb.1: +; X86-BMI1NOTBM-NEXT: movl %eax, %edi +; X86-BMI1NOTBM-NEXT: .LBB37_2: +; X86-BMI1NOTBM-NEXT: movl %edi, %eax +; X86-BMI1NOTBM-NEXT: shrl %cl, %eax +; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: movl $0, %edx +; X86-BMI1NOTBM-NEXT: jne .LBB37_4 +; X86-BMI1NOTBM-NEXT: # %bb.3: +; X86-BMI1NOTBM-NEXT: movl %esi, %ebx +; X86-BMI1NOTBM-NEXT: movl %eax, %edx +; X86-BMI1NOTBM-NEXT: .LBB37_4: +; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %ebx +; X86-BMI1NOTBM-NEXT: testb $32, %cl +; X86-BMI1NOTBM-NEXT: jne .LBB37_6 +; X86-BMI1NOTBM-NEXT: # %bb.5: +; X86-BMI1NOTBM-NEXT: movl %ebx, %eax +; X86-BMI1NOTBM-NEXT: .LBB37_6: +; X86-BMI1NOTBM-NEXT: popl %esi +; X86-BMI1NOTBM-NEXT: popl %edi +; X86-BMI1NOTBM-NEXT: popl %ebx +; X86-BMI1NOTBM-NEXT: retl +; ; X86-BMI1BMI2-LABEL: bzhi64_d3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %edi @@ -2052,6 +3047,16 @@ define i64 @bzhi64_d3_load_indexzext(i64* %w, i8 %numlowbits) nounwind { ; X64-NOBMI-NEXT: shrq %cl, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_d3_load_indexzext: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl %esi, %ecx +; X64-BMI1NOTBM-NEXT: movq (%rdi), %rax +; X64-BMI1NOTBM-NEXT: negb %cl +; X64-BMI1NOTBM-NEXT: shlq %cl, %rax +; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx +; X64-BMI1NOTBM-NEXT: shrq %cl, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1BMI2-LABEL: bzhi64_d3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi @@ -2189,6 +3194,12 @@ define i64 @bzhi64_constant_mask64(i64 %val) nounwind { ; X64-NOBMI-NEXT: andq %rdi, %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_constant_mask64: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl $15872, %eax # imm = 0x3E00 +; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1TBM-LABEL: bzhi64_constant_mask64: ; X64-BMI1TBM: # %bb.0: ; X64-BMI1TBM-NEXT: bextrq $15872, %rdi, %rax # imm = 0x3E00 @@ -2218,6 +3229,12 @@ define i64 @bzhi64_constant_mask64_load(i64* %val) nounwind { ; X64-NOBMI-NEXT: andq (%rdi), %rax ; X64-NOBMI-NEXT: retq ; +; X64-BMI1NOTBM-LABEL: bzhi64_constant_mask64_load: +; X64-BMI1NOTBM: # %bb.0: +; X64-BMI1NOTBM-NEXT: movl $15872, %eax # imm = 0x3E00 +; X64-BMI1NOTBM-NEXT: bextrq %rax, (%rdi), %rax +; X64-BMI1NOTBM-NEXT: retq +; ; X64-BMI1TBM-LABEL: bzhi64_constant_mask64_load: ; X64-BMI1TBM: # %bb.0: ; X64-BMI1TBM-NEXT: bextrq $15872, (%rdi), %rax # imm = 0x3E00 |